Scholarly Literature Research

Scholarly Literature Research

Use this skill as the default path for the user's literature-review and reference-refresh work, especially when searching WOS/Web of Science and IEEE Xplore.

Core Workflow

1. Anchor the search to the manuscript context.
   • Identify the old reference, the sentence or paragraph it supports, and the claim that a replacement must support.
   • Generate several concise keyword queries from that claim.
2. Search authoritative databases.
   • Use playwright-browser-automation for visible login, institution authentication, and browser session management.
   • Prefer non-visual automation with playwright-cli run-code / eval and the scripts in this skill.
   • Use screenshots only for QA or when the page structure is broken.
3. Extract structured candidates.
   • Save JSON outputs with title, year, source hints, DOI hints, link, and result snippet.
   • Deduplicate by title and group candidates by the old reference they could replace.
4. Screen lightly before asking the user to download PDFs.
   • Favor recent papers, high-quality journals/conferences, relevant methods, and direct fit to the manuscript sentence.
   • Report titles first; do not overclaim relevance before reading full PDFs.
5. After PDFs are available, use pdf-plus-mineru for deeper reading and relevance validation.
6. For manuscript insertion or formatting, use documents and preserve Zotero/Word formatting rules.
7. For final PDF reference validation, apply pdf-plus-mineru two-column safeguards.
   • Render and inspect the reference pages when the manuscript uses IEEE-style two-column layout.
   • Do not treat text-only extraction order as authoritative for missing or reordered references.

Local Scripts

Run from the task workspace. By default the bundled scripts write temp files under the current working directory.

Extract the current results page:

node scripts/extract_results.js --session ieee --site ieee --out .\output\literature-search\ieee_current.json

Run one search and extract:

node scripts/run_search.js --site ieee --session ieee --query "robot calibration review kinematic calibration survey" --out .\output\literature-search\ieee_ref10.json

Run the first query for each ICMA reference-refresh target:

node scripts/batch_icma_search.js --site ieee --session ieee --limit-queries 1

Database Notes

• IEEE Xplore is validated for direct DOM extraction on this machine.
• WOS/Web of Science login and basic-search navigation are validated, but WOS title extraction may require selector tuning on a fresh result page.
• Avoid parallel navigation and extraction in the same browser session; navigation can destroy the page execution context.
• Complex WOS Boolean expressions are fragile. Start with natural-language topic queries, then refine once results are visible.

References

• For the current ICMA manuscript replacement task, use references/queries_icma_refs.json.
• This skill expects playwright-browser-automation to also be installed in the user's Codex skills directory.